Joint learning of feature and topology for multi-view graph convolutional network.

Area of Science:

• Machine Learning
• Computer Vision
• Data Science

Background:

• Graph convolutional networks (GCNs) are widely used for semi-supervised classification.
• Existing GCNs often treat multi-view data features and topology separately, limiting performance.
• This leads to underutilization of multi-view data's consistency and complementarity.

Purpose of the Study:

• To propose an end-to-end joint fusion framework for multi-view semi-supervised classification.
• To simultaneously achieve consistent feature integration and adaptive topology adjustment.
• To leverage the consistency and complementarity of multi-view data for improved intrinsic information capture.

Main Methods:

• A deep matrix decomposition module for consistent feature representation across views.
• A flexible graph convolution for adaptive and robust topology learning.
• A joint fusion framework where feature and topology modules mutually enhance each other.

Main Results:

• The proposed method effectively integrates features from multiple views.
• Adaptive topology adjustment reduces the impact of unreliable information.
• Experimental results demonstrate superior performance compared to state-of-the-art semi-supervised classification methods.

Conclusions:

• The joint fusion framework successfully captures intrinsic information from multi-view data.
• Simultaneous feature consistency and topology adaptivity are key to improved classification.
• This approach offers a more effective way to utilize multi-view data in GCNs.